[goog00]

Summary

• Registers asinf32 / asinf64 / acosf32 / acosf64 placeholder symbols in dialect-mir/src/rust_intrinsics.rs, following the same pattern as the existing float math intrinsics
• Recognizes both call shapes in the mir-importer float-math dispatch:
  • the std inherent methods f{32,64}::asin/acos, which lower through the std::sys::cmath::asinf / asin / acosf / acos shims (asin/acos are std-only — there is no core libm form, unlike cbrt)
  • the libm-crate spellings (libm::asinf, libm::math::acos::acos, ...) that glam's nostd-libm lowering emits on nvptx
• Maps them to CUDA libdevice builtins __nv_asinf / __nv_asin / __nv_acosf / __nv_acos in the MIR-lower call translator
• Hooks the new placeholders into the codegen collector's cmath-shim allowlist
• Adds an asin_acos_smoke example: 14 test values across the full [-1, 1] domain (including both endpoints and -0.0) × 4 variants (asin/acos × f32/f64), verifying GPU results are within 2 ULP of host libm

Test plan

• cargo test -p dialect-mir -p mir-importer -p mir-lower --lib — unit tests covering intrinsic name → enum dispatch, the std::sys::cmath + libm-crate spellings, and arg count
• cargo oxide run asin_acos_smoke — end-to-end GPU smoke test; expected output: SUCCESS: 14 cases × 4 variants within 2 ULP of host libm
• Verified on RTX 5090 (sm_120a, CUDA 13.0)

[nihalpasham]

Thanks for the PR. LGTM

The libdevice routing is correct and the is_libm_path guard prevents false-positive matches.

Note this slightly widens the blast radius of issue #98 (NVVM IR on pre-Blackwell targets), but that's the right layering - the intrinsics should land independently of the NVVM IR fix.